Intel has launched its new generation of Core i5/i7 processors, codenamed Haswell. If you are familiar with Intel’s tick-tock strategy, this is a ‘tock’: a microarchitecture update on existing process technology. In this case the 22 nm process which is also used for its predecessor, codenamed Ivy Bridge.
For all the details on the new CPU internals and its performance, I would like to direct you to the many review sites that cover the new CPU extensively, such as Anandtech. I already covered Haswell in an earlier blog as well. I do not have that much to add to that, other than to keep in mind that the benchmarks of the review sites will most probably not include any software that takes advantage of TSX yet, so we won’t see the true value of that new addition until later, when it is widely adopted in software.
Other than that, the CPU is not all that spectacular. It does raise the bar in performance somewhat, and it does have some improved power saving features on board, but that’s about it. What IS spectacular though, is the integrated GPU.
Anandtech decided to do a separate article on the new iGPU, codenamed Iris Pro. This is the big improvement in Haswell. Intel has always had GPUs that were quite energy-efficient, but this time, they actually have quite respectable performance:
The Iris Pro 5200 can actually get close to a GeForce GT 650M in performance, while consuming only about 60% of the power. Compared to AMD’s APUs, things get even more interesting. AMD’s fastest mobile part, the A10-4600M is no match for the Iris Pro. In fact, even the desktop part A10-5800K (rated at 100W!) can’t quite reach the level of performance of the Iris Pro.
Also good to see that unlike AMD, Intel has no tessellation issues either:
It is quite impressive in certain OpenCL workloads as well:
All in all, it seems that Intel has closed the gap with AMD’s APUs, and is threatening nVidia’s lower-end GPUs as well. I think especially AMD’s APUs will have a hard time dealing with Haswell. As I already said earlier on Trinity, AMD has the problem of being CPU-limited. Trinity already had a considerably faster GPU than Llano on board, but because the CPU performance was virtually the same, the real-world gains of the faster GPU were modest. Even back then, Intel’s iGPUs could already outperform Trinity in some of the more CPU-intensive games.
AMD’s APUs had only one advantage, and that was the faster GPU. Intel has now taken that advantage away, and it will be very difficult for AMD to regain that advantage. AMD does not have much room to move forward at this point. Putting an even faster GPU on the same Piledriver-based quadcore CPU is going to yield little or no real-world gains. They would have to improve the CPU-portion. But how will they do that? If you see that even the 100W desktop part A10 is already trailing considerably behind even a 47W Haswell… Things look bleak for AMD. They would need to massively improve CPU performance while at the same time cutting power consumption in half, before they can get into the markets that Haswell is aiming at.
I would say that Haswell is APU/Fusion done right: no compromise on the CPU-part, while at the same time the iGPU is fast enough for some light gaming, and the iGPU can also boost applications that support DirectCompute, OpenCL or QuickSync. All that in a very modest power envelope.